A status selection planning system is one of planning expert systems. In this system, the most promising status from tentative statuses generated by applying the dispatching rules is selected by a status selection rule. Dispatching rules mean fragmentary and convenient assignment algorithms. Quality of the solution depends on the knowledge-base. However, it is usually difficult to acquire useful knowledge from human experts.
In this paper, a learning method of a status selection rule set using GA (Genetic Algorithm) is proposed. The status selection rule set is regarded as an individual. The representation of scheduling knowledge by gene is generally difficult in GA, because it is necessary to carry out crossover operation on gene. To cope with the representation problem, a status selection rule is represented by tree construction and a status selection rule set is represented by a list of those tree construction. From the results of the application to a simple job shop problem, it is shown that the knowledge acquired by the proposed method is superior to the human's knowledge